Follow up continuation audit retention and Context v1 ergonomics

[marcusglee11]

Context

Follow-up from completed LifeOS issue #100 and the continuation hook audit retention rollout.

Closed proof:

• Issue Add retention for continuation hook audit log #100: Add retention for continuation hook audit log #100
• Hermes Agent implementation PR: https://github.com/marcusglee11/hermes-agent/pull/11
• Hermes Agent merge commit: c5c71e981b855c2757f4ef7ef2e7aa1d74feb63a
• LifeOS tracking PR: Track continuation audit retention workstream #105
• LifeOS merge commit: 9b0ed4342b865fe3b6f8b10d081525750fd059b3

Problem

Issue #100 shipped bounded max-record retention for continuation hook audit JSONL logs. AA approved it with non-blocking follow-ups. The work also dogfooded Context Management v1 and exposed ergonomics gaps.

Follow-up scope

Keep this as a follow-up/control issue. Do not reopen #100.

Hermes Agent runtime audit hardening

• Add optional byte or per-field-size guardrails for audit records, especially evidence, new_backlog, and prompt-like fields.
• Clarify native Windows/non-POSIX concurrency semantics, or add a cross-platform lock implementation.
• Decide whether HERMES_WORKSTREAM_AUDIT_LOG_MAX_RECORDS=0 / negative values should remain an intentional unbounded escape hatch or be treated as invalid and fall back to the default 2000.

Context Management v1 ergonomics

• Add mutation commands for common state changes:
  • status / lifecycle_state
  • preflight PASS/FAIL/UNKNOWN
  • refs/evidence refs
  • phase / next_action
• Add first-class multi-repo/external implementation refs so a LifeOS issue can cleanly track a Hermes Agent implementation PR without stuffing external state into generic refs.

Acceptance criteria

• Follow-up implementation is split into issue-sized PRs; this issue may close once child issues/PRs are created or the above scope is explicitly dispositioned.
• No change treats state.yaml as completion truth. Completion truth remains PR/CI/main-readback/issue receipt.
• No background scheduler/cron cleanup is introduced unless a later issue explicitly approves it.
• Any Hermes Agent runtime change includes focused tests and local/runtime readback.
• Any Context v1 schema/tooling change validates existing workstream states or includes a migration plan.

Notes

Priority: useful but not urgent. The v0 retention implementation is operationally acceptable as shipped.